PixelDiT Family: VAE-Free Pixel-Space DiT + PiD Super Resolution by NVIDIA

PixelDiT by NVIDIA is a VAE-free pixel-space diffusion transformer for text-to-image generation. Includes PiD (Pixel Diffusion Decoder) models for super-resolution up to 4K.

P

PixelDiT by NVIDIA

Text-to-ImageVAE-freeSuper ResolutionDiT

PixelDiT is a VAE-free pixel-space diffusion transformer (1.3B) for text-to-image generation, published at CVPR 2026. This family also includes PiD (Pixel Diffusion Decoder) — 4-step distilled decoders that upscale latent images from 512px to 2048px or 1024px to 4096px across multiple latent spaces (FLUX, SD3, SDXL).

HF OrgGitHubPaperWebsite

Components

ComponentDescription
PixelDiT & PiDPixelDiT 1.3B text-to-image + PiD super-resolution decoders for multiple latent spaces

Select a component for tutorials, custom nodes, and related news. Curated weight downloads live in the Model Hub catalog.